5 min read · 14 sources

A developer can burn through a thousand dollars of tokens in hours, making software potentially more expensive.

Slackbot Now Builds Live Data Surfaces From a Sentence

Source: slack.com ↗

Slack dropped its AI-powered interface for work, and it’s the biggest story of the day because it changes where analytics actually happen. Slackbot can now turn natural language requests into live, interactive “Slackforce Surfaces” - think filtering, exploring, and acting on data without leaving the chat window. For engineers, this means the BI tool you spent months wiring up is now something a sales manager can drive from a thread. It doesn’t replace your warehouse; it replaces the friction between asking a question and seeing the answer. If you run a Slack workspace, expect a wave of “can the bot do this?” requests.

Emergency Windows 11 Update Patches RDS, But AMD GPU Errors Persist

Source: windowslatest.com ↗

Microsoft pushed out-of-band update KB5129195 after the September 2026 Patch Tuesday (KB5124008) broke things. The emergency build advances 25H2 to Build 26200.9457 and 24H2 to Build 26100.9457, fixing Remote Desktop Services and security holes. But read the fine print: it does not address AMD GPU errors. If you’re on affected hardware, you’re still stuck with the instability. Engineers should install this to restore RDS and close the security gaps, but budget for a separate AMD driver investigation or you’ll be chasing the same crash in a week.

AI Was Supposed to Make Software Cheaper. It's Not.

Source: galratner.substack.com ↗

Gal Ratner’s essay dismantles the cost-plus fallacy that’s been driving procurement decisions. Software has never been priced on production cost - the second copy of Windows cost nothing to manufacture, and it still sold for hundreds of dollars. With AI, a developer can burn through a thousand dollars of tokens in hours, making software potentially more expensive to build. The takeaway for anyone running an engineering org: AI changes your cost structure, but it doesn’t change how buyers value the output. Expect budgets to go up, not down, and plan your token spend like you plan your cloud bill - because it’s now a line item.

When Your CRM Starts Building Itself

Source: bcg.com ↗

BCG’s piece on the agentic SaaS era follows the Salesforce Dreamforce narrative where customers are using more of the platform since AI agents arrived. The catch: business users are now building apps faster than IT can govern them. CIOs are told to stop resisting and instead own the foundations - identity, data architecture, observability, integration - so the business can build safely on top. For engineers, this is a preview of the shadow IT problem on steroids. If you’re not preparing governance rails for business-led development, you’re going to be the one debugging a production incident caused by a no-code agent someone shipped on a Friday.

Westpac's Adapt: 1PB and 14,000 Pipelines on Azure

Source: itnews.com.au ↗

Westpac unveiled Adapt, an enterprise data platform on Microsoft Azure that consolidates data from 285 source systems into a single governed environment. The migration moved over 1PB of data and more than 14,000 data pipelines, completing in March 2026. It’s a reference architecture for anyone doing a large-scale data consolidation: one governed environment, with data at rest where it lives rather than moving it. The scale matters - if a bank can do this with 285 systems, your org’s “we’re too complex” excuse just died.

Salesforce Builder Central: Apps From a Conversation

Source: salesforce.com ↗

Salesforce’s Builder Central lets admins and business users build and configure apps and AI agents using plain language. It’s scheduled to launch the week of September 21. Instead of dragging fields around a form, you describe what you want and the platform generates a working app. For engineers, this is the conversational interface trend applied to app creation itself. It doesn’t replace your code, but it does mean the backlog of “small internal tools” is about to vanish, replaced by a new backlog of “who owns this agent and what can it actually do?”

HubSpot Tightens App Governance With Least-Privilege Scopes

Source: developers.hubspot.com ↗

HubSpot’s Fall 2026 release is all about governance. Broad OAuth scopes are being replaced with least-privilege equivalents - think crm.objects.tickets.read instead of .write - effective September 8, 2026. Access now stops at install time if a user lacks permissions, and admins can require approval for new connectors, restrict who can install them, and see who owns each integration. There’s even a warning when deactivating someone who owns an integration. This is the pattern to copy if you’re building a SaaS platform: per-user spend visibility, usage reporting, and granular scopes are table stakes now, not features.

Zoom's Revenue OS: AI Connecting the Dots

Source: news.zoom.com ↗

Zoom unveiled its AI-powered Revenue OS, connecting signals from first contact through pipeline generation so AI can recommend next actions. The pitch is that conversation data and buyer signals now live together, improving go-to-market efficiency. The article lacks specifics on APIs or infrastructure, so take the integration claims with a grain of salt until there’s a technical deep dive. For engineers, the takeaway is that the CRM is no longer a system of record - it’s becoming a system of suggestion.

From Obsidian Vault to Enterprise Company Brain

Source: ssp.sh ↗

This piece walks through scaling a personal Obsidian vault to an enterprise “company brain” where AI agents write, branch, and merge knowledge with versioning and verification. It recommends open formats like Markdown for longevity and suggests agents write on separate branches to avoid collisions, with human review governing merges. The catch: single-user graph systems don’t scale to multi-agent writes. If you’re building AI-driven knowledge management, you need enterprise-grade versioning and merge controls, not a shared folder.

JFrog Artifactory: Three Bugs Under Active Attack

Source: devops.com ↗

Security researchers at Wiz report active exploitation of three JFrog Artifactory vulnerabilities. Two of them - CVE-2026-42018 (improper authentication returning anonymous tokens) and CVE-2026-42016 (privilege escalation via token-scope validation) - chain together to go from low privilege to full admin. All three have patches. If you’re running Artifactory and haven’t applied the latest updates, this is your queue. Check your access logs for anonymous token usage and audit privilege escalations now, not after the weekend.

Quick Hits

Source: spectrocloud.com ↗

• Spectro Cloud’s AMD Instinct Coder delivers GPUs with the inference stack pre-flashed, cutting months of setup to a day.
• Microsoft Teams now lets external presenters manage event production controls, with GA rolling out through early August.
• Anthropic’s webinar on scaling Claude with cost controls covers per-teammate spend visibility and usage tracking, configurable by one admin in a day.
• AI cloud bills are forcing leaders to separate GPU spend from general cloud commitments and negotiate more flexible capacity terms.